1. Field of the Invention
The present invention relates to a design technique for a semiconductor integrated circuit. In particular, the present invention relates to a design technique in which layout around an element is considered.
2. Description of Related Art
Characteristics of a transistor depend on “device parameters” such as gate length L, gate width W, mobility μ, threshold voltage Vt and the like. When the transistor is actually manufactured, those device parameters are likely to vary from design values. The variations of the device parameters from the design values lead to variations of the transistor characteristics from expected values. It is therefore important to sufficiently consider variation factors of the device parameters at circuit design, verification and manufacturing stages.
For example, a typical variation factor of a gate size (L and W) is OPE (Optical Proximity Effect) in photolithography processes. Due to the OPE, a resist pattern may not be formed as intended, which causes deviations of interconnection width and gate size from the design values. In particular, the variation from the design value caused by the OPE becomes more conspicuous with increasing miniaturization of semiconductor manufacturing processes. Also, the OPE becomes remarkable in a case of a high pattern density. Therefore, OPC (Optical Proximity Correction) is generally performed, in which the OPE is estimated to correct a mask data in advance (refer to US Publication 2007/0033558, for example).
Japanese Laid-Open Patent Application JP-2003-23080 discloses a technique to change a buffer cell depending on cell density in order to obtain a desired chip speed. More specifically, plural types of buffer cells having different driving capability are first prepared, and then driving capability of a buffer cell is changed at a layout design stage depending on the cell density.
Moreover, a layout pattern (or structure) surrounding a transistor can affect device parameters and characteristics of the transistor. Such a peripheral factor affecting the characteristics includes: (1) gate pitch; (2) well proximity effect; and (3) STI stress, for example.
(1) The gate pitch is an interval between a gate of a transistor and a gate of the adjacent transistor. The gate pitch affects, for example, ion dose in an ion injection process. It is known that ON-current becomes larger as the gate pitch becomes smaller while ON-current becomes smaller as the gate pitch becomes larger.
(2) The well proximity effect is described, for example, in Japanese Laid-Open Patent Application JP-2007-36249. More specifically, in a well ion injection process, a part of the injection ions collide with an edge of a resist pattern to be scattered. The scattered ions intrude into an active region of the transistor, which varies the threshold voltage Vt from the design value. This is the well proximity effect. The influence of the well proximity effect becomes larger, as an interval between an edge of the well and an edge of the active region becomes smaller.
(3) The STI stress, which is stress imposed on a transistor from an STI (Shallow Trench Isolation) that is a device isolation structure, affects the mobility μ in a channel (refer to Japanese Laid-Open Patent Application JP-2006-178907) When a width of the STI becomes narrower, the STI stress is increased, and thereby the mobility μ and the ON-current are changed.
As described above, a peripheral pattern (peripheral structure) around a transistor causes variation of the transistor characteristics from the design values. That is to say, the transistor characteristics have “peripheral pattern dependence”. If the peripheral pattern is different between transistors, the characteristics of the transistors after manufacturing may become different, even if the transistors have the same design. It is therefore important to consider the peripheral pattern dependence of transistor at the semiconductor integrated circuit design and verification stages.
Japanese Laid-Open Patent Application JP-2006-178907 discloses a method of circuit simulation in which the STI width is considered. According to the circuit simulation, the mobility μ which is one of model parameters of a transistor is corrected depending on the STI width. More specifically, an approximate expression that expresses a relationship between the STI width and the mobility μ is created beforehand based on actually measured values or the like. Prior to the circuit simulation, a designed STI width (design value) is extracted based on the circuit layout. Then, the mobility μ as the model parameter is corrected by using the extracted STI width and the above-mentioned approximate expression. After that, the circuit simulation is performed by using the corrected model parameter. Consequently, the transistor characteristics dependent on the STI width are considered and accuracy of the circuit simulation is improved.
In this context, a tool is commercially available that modulates (corrects) a SPICE netlist used in a SPICE simulation in consideration of the peripheral pattern dependence. The tool is hereinafter referred to as “pattern dependence consideration tool”.
FIG. 1 is a schematic diagram for explaining a function of the pattern dependence consideration tool. First, a model (e.g. Δμ=f(X)) that expresses a relationship between graphical information (e.g. STI width X) and modulation amount (e.g. Δμ) of a SPICE model parameter is created through evaluation of TEG (Test Element Group). After that, the SPICE simulation of a cell is performed.
Prior to the SPICE simulation, the pattern dependence consideration tool reads a SPICE netlist (pre-modulation cell netlist) of a target cell and a layout data (cell layout data) of the target cell. Subsequently, the pattern dependence consideration tool extracts the graphical information (X) from the cell layout data, and rewrites (modulates) the SPICE model parameter included in the pre-modulation cell netlist with reference to the above-mentioned model (Δμ=f(X)). As a result, a post-modulation cell netlist that is a SPICE netlist after the rewriting is generated. In the SPICE simulation, the post-modulation cell netlist output from the pattern dependence consideration tool is used. Since the peripheral pattern dependence of transistor is reflected in the SPICE netlist, the simulation accuracy is dramatically improved. It should be noted that the SPICE model parameter to be modulated is not limited to the mobility μ but can include the gate length L, the gate width W, the threshold voltage Vt or the like.
Next, a method of designing and verifying a semiconductor integrated circuit by utilizing the pattern dependence consideration tool will be described with reference to FIG. 2. The pattern dependence consideration tool is used at a cell characterization stage. As mentioned above, the pattern dependence consideration tool rewrites the pre-modulation cell netlist based on the cell layout data to generate the post-modulation cell netlist. After that, the SPICE simulation is performed by using the post-modulation cell netlist, and a delay value (expected value) of the cell is calculated. At this time, a gate size and the like of the transistor are set to corner values in consideration of manufacturing variability. As a result, cell delay values corresponding to the manufacturing variability can be obtained. A corner delay value is extracted from the calculated cell delay values, and a cell delay library that provides the corner delay value is generated. In this manner, the cell characterization is performed.
At the design/verification stage, a placement and routing tool performs cell placement and routing based on a netlist of a design circuit. As a result, a layout data indicating layout of the design circuit is generated. Next, a delay calculation tool reads the netlist, the generated layout data and the above-mentioned cell delay library, and calculates delay values in the design circuit. More specifically, the delay calculation tool calculates delay values of inter-cell interconnections, based on RC extracted from the layout data. The cell delay values (corner delay values) of cells are obtained from the cell delay library. Thus, a delay file indicating the cell delay values and path delay values is generated.
Next, delay verification (timing verification) of the design circuit is performed. More specifically, the STA tool performs an STA (Static Timing Analysis) based on the netlist and the generated delay file. If the verification result is “FAIL”, the above-described placement and routing process is performed again to modify the layout data. The same processing is repeated until the verification result becomes “PASS”.
With regard to the method shown in FIG. 2, there are the following problems. FIG. 3 is a schematic diagram for explaining the problems, showing one example of a cell layout used at the characterization stage. As an example, let us consider a primitive cell such as a NAND gate, an inverter or the like. The cell shown in FIG. 3 has PMOS transistors P1 to P3 and NMOS transistors N1 to N4 which are formed in a region sandwiched between a power supply line VDD and a ground line GND. The PMOS transistor P1 has diffusion regions PD1 and PD2 that are in contact with an STI. The PMOS transistors P2 has diffusion regions PD3 and PD4 that are in contact with the STI. A distance between diffusion regions of the respective PMOS transistors P1 and P2 that are placed adjacent to each other, namely, a distance X2 between the diffusion regions PD2 and PD3 is the “STI width”.
According to the method shown in FIG. 2, the pattern dependence consideration tool extracts the graphical information (e.g. STI width X) from the cell layout data to calculate the modulation amount (e.g. Δμ) of the SPICE model parameter. As to the PMOS transistor P2 shown in FIG. 3, the peripheral layout pattern (e.g. STI width X2) is already known and thus the SPICE model parameter can be modulated. As to the PMOS transistor P1 placed at the outermost of the cell, on the contrary, not all peripheral layout pattern is known at the characterization stage where surrounding cells (cells on the left and right sides) are not placed yet. In FIG. 3, for example, an STI width X1 on the left side of the PMOS transistor P1 is not known at the characterization stage but becomes known at the subsequent cell placement process. Therefore, it is not possible to module the SPICE model parameter of the PMOS transistor P1 prior to the SPICE simulation. The same applies to the PMOS transistor P3, the NMOS transistors N1 and N4.
In this manner, with regard to a transistor placed at an inward area of the cell, it is possible by the above-described method to consider the peripheral pattern dependence in the SPICE simulation. However, with regard to the outermost transistor placed near the cell boundary, it is not possible to consider the peripheral pattern dependence in the SPICE simulation since the peripheral layout pattern is unknown. In other words, the peripheral pattern dependence cannot be fully considered at least with respect to the characteristics of the outermost transistor. Since the transistor characteristics affect the cell delay value, it cannot be said that the peripheral pattern dependence is fully considered in the generated cell delay library. Consequently, it cannot be said that the accuracy of the delay verification (STA) is sufficient. This contributes to reduction of yield and reliability of the semiconductor integrated circuit to be manufactured.
In view of the above-described problems, the following two new methods are suggested as alternatives to the method shown in FIG. 2.
According to the first method, types of cells adjacent to a target cell are assumed at the characterization stage for the target cell. The pattern dependence consideration tool modulates the cell netlist based on layout data of the assumed adjacent cells in addition to the cell layout data of the target cell. The resultant post-modulation cell netlist is used in the SPICE simulation so that the cell delay value of the target cell is calculated. The same processing is repeated with setting the adjacent cell type variously (for each of prepared types). As a result, plural kinds of cell delay values corresponding to respective combinations of the adjacent cells are obtained. Then, the cell delay library is generated based on the obtained plural kinds of cell delay values. The peripheral pattern dependence of the outermost transistors in the cell is considered in the generated cell delay library. However, the time (TAT) required for the cell characterization is increased as compared with that in the method shown in FIG. 2.
Furthermore, the first method is disadvantageous in that the corner condition of the cell delay value becomes stricter as compared with that in the method shown in FIG. 2. FIG. 4 schematically shows the corner condition (corner delay value) provided by the cell delay library. According to the method shown in FIG. 2, a delay value distribution corresponding to only the normal manufacturing variability is obtained. According to the first method, on the contrary, the various types of adjacent cells are assumed and thus a plurality of delay value distributions having different center delay values are obtained. Therefore, a variability range of the delay value of each cell is expanded as compared with the case where the adjacent cells are not assumed. That is to say, a “margin” corresponding to the difference of the combination of adjacent cells is further added to the variability range of the delay value of each cell. This means that the corner condition (corner delay value) becomes stricter.
At the circuit design/verification stage, it is necessary to perform the layout design such that the design circuit can operate even in the corner condition. In other words, it is necessary to repeat the layout design and delay verification until the design circuit operates in the corner condition. Therefore, when the corner condition becomes stricter, the number of the repetition times is inevitably increased and thus the time (TAT) required for the circuit design/verification is increased. Moreover, since the margin is added to the cell delay value such that every adjacent cell types can be supported, it is necessary to increase driving capability of transistors and/or to additionally insert redundant cells. This leads to increase in a chip area and power consumption. According to the first method, as described above, the design/verification TAT, chip area and power consumption are increased although the peripheral pattern dependence is fully considered.
According to the second method, the cell characterization is performed as in the case of the method shown in FIG. 2. As a result, the cell delay library in which the peripheral pattern dependence is partially considered is generated. After that, the pattern dependence consideration tool is utilized again at the circuit design/verification stage. More specifically, the pattern dependence consideration tool modulates the SPICE netlist of each cell with reference to the chip-level layout data generated in the placement and routing process. Subsequently, the SPICE simulation of each cell is performed again by using the post-modulation netlist, and a “difference (variation)” from the cell delay value given by the cell delay library is calculated. The calculated difference is reflected in the delay file. In this manner, the peripheral pattern dependence can be considered with regard to the outermost transistor in the cell.
The same processing is repeated for all cells (for example, millions of cells) included in the design circuit. That is to say, the modulation of the SPICE netlist and the SPICE simulation are executed for millions of times at the cell design/verification stage. Thus, the design/verification TAT is increased also in the second method, although the peripheral pattern dependence is fully considered.
According to the related techniques, as described above, while the peripheral pattern dependence is considered, the design/verification TAT is increased. A technique is desired which can prevent the increase in the design/verification TAT for the semiconductor integrated circuit, while considering the peripheral pattern dependence of the transistor characteristics sufficiently.